On 25 November 2014 at 18:58, David Vasek <va...@fido.cz> wrote:
> did not look neither efficient, nor healthy. Try dd if=/dev/zero > of=/dev/rsd1c bs=1m while watching systat/iostat at the same time. Is it > still the case? So here are the findings. The test is virtualised but below is the baseline into a vnd container (no crypt etc) GENERIC 5.6-current amd64: # dd if=/dev/zero of=/dev/rvnd0a bs=1m count=1000 1000+0 records in 1000+0 records out 1048576000 bytes transferred in 41.652 secs (25174524 bytes/sec) The results for vnd container using crypt (noticed ~20% cpu utilisation): # dd if=/dev/zero of=/dev/rvnd0a bs=1m count=1000 1000+0 records in 1000+0 records out 1048576000 bytes transferred in 59.270 secs (17691307 bytes/sec) Crypto softraid inside a vnd container (noticed ~8% cpu utilisation): # dd if=/dev/zero of=/dev/rsd2a bs=1m count=1000 1000+0 records in 1000+0 records out 1048576000 bytes transferred in 87.422 secs (11994340 bytes/sec) I'll continue using vnd crypt container based on advice from theo@ . Performance is not a real issue for this small amount of data that is not accessed often in my use case so I wasn't too concerned either way, just wanted to make sure I was on a valid supported path.
